Google's internal-only coding AI 'GfG': Full details, trained on 1 trillion tokens of internal data using Gemini, developed to halve task completion time
A paper by Google researchers introduces the practical effects of 'GfG', an internal coding AI built by training Gemini on 1 trillion tokens of internal data.
Key points
- •Researchers at Google published the paper 'Customizing an LLM for Enterprise Software Engineering' in October
- •The study details a method for training AI on massive proprietary development data to build coding models specialized for internal system development
- •The report indicates that the developed model can halve the time required for certain tasks